MDA and PCA projection subspaces for electrophysiological data set.

<p>A) First MDA dimension separates all startles from the rest data points (black dots), while the second dimension separates shake (diamonds) events from the metal sound events (circles). Training and test data points are represented in black and red color. Two dimensional gaussian distribution are fitted to the projected points for each class, revealing that they form segregated clusters (ellipsoids extend up to the 2<i>σ</i> boundaries). The colors black, green, blue, magenta and cyan denote rest, metal sound, air, drop and shake events, respectively. B) The third MDA dimension separates shake and airblow (triangles) classes from drop class (stars), and the fourth dimension separates metal sound and shake classes from drop and air blow classes. C) Inspection of the PCA discriminant dimensions indicate a similar structure of the encoding subspace with the first dimension separating the base from the rest of the startles and the second dimension separating the metal sound from shake. D) The third PCA dimension separates air blow from drop and the last dimension PCA separates shake from drop.